Perbandingan Kinerja Metode Machine Learning Dan Deep Learning Untuk Klasifikasi Multilabel Jenis Emosi

Pratama, Fauzi Rizki (2025) Perbandingan Kinerja Metode Machine Learning Dan Deep Learning Untuk Klasifikasi Multilabel Jenis Emosi. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5025211220-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (9MB) | Request a copy

Abstract

Perkembangan teknologi digital telah membawa dampak signifikan dalam berbagai aspek kehidupan, termasuk dalam cara manusia berinteraksi dan berkomunikasi melalui media sosial. Platform seperti Facebook, Instagram, Twitter, dan Reddit telah menjadi ruang bagi individu untuk berbagi emosi dan informasi, namun dampaknya terhadap kesejahteraan emosional, terutama dalam hal depresi, semakin menjadi perhatian. Penelitian ini bertujuan untuk menganalisis penggunaan model Artificial Intelligence (AI), khususnya pemrosesan Natural Language Processing dan Deep Learning, dalam mendeteksi emosi negatif pada pengguna media sosial, dengan fokus pada klasifikasi jenis emosi negatif. Metode yang digunakan dalam penelitian ini melibatkan pendekatan Machine Learning, yaitu Support Vector Machine (SVM), Extreme Gradient Boosting (XGBoost), Light Gradient Boosting Machine (LIGHT GBM), dan Categorical Boosting (CatBoost). Metode lain yang digunakan dalam penelitian ini juga melibatkan pendekatan Deep Learning, yaitu Bidirectional Encoder Representations from Transformers (BERT), Decoding-enhanced BERT with Disentangled Attention (DistilBERT), Robustly Optimized BERT Approach (RoBERTa), dan juga Decoding-enhanced BERT with Disentangled Attention (DeBERTa) yang memiliki keunggulan dalam mengidentifikasi emosi secara otomatis dari teks yang diunggah pengguna. Dataset yang digunakan adalah DepressionEmo, yang dirancang untuk mendeteksi 8 emosi yang terkait dengan depresi seperti Anger, Cognitive dysfunction, Emptiness, Hopelessness, Loneliness, Sadness, Suicide intent, dan Worthlessness melalui 6037 contoh posting dengan teks yang panjang oleh pengguna Reddit. Hasil penelitian ini diharapkan dapat menghasilkan model yang efektif untuk mengklasifikasi jenis emosi berdasarkan teks, memberikan wawasan yang lebih mendalam mengenai jenis emosi negatif, dan menawarkan solusi praktis dalam pengembangan model AI untuk mendeteksi emosi, dan memberikan evaluasi terkait model-model AI dalam mendeteksi emosi. Variasi Deep Learning terbaik yang digunakan adalah microsoft/deberta-base dengan hyperparameter berupa hidden layers bernilai 12, batch size bernilai 8, learning rate bernilai 2e-5, dan jenis optimizer Adam yang memperoleh nilai f1 micro 0,8254. Perbandingan metode Machine Learning juga menunjukkan model deep learning memberikan nilai yang lebih baik pada metriks evaluasi F1 Micro Score.
=========================================================================================================================================
The development of digital technology has had a significant impact on various aspects of life, including how people interact and communicate through social media. Platforms such as Facebook, Instagram, Twitter, and Reddit have become spaces for individuals to share emotions and information; however, their impact on emotional well-being, particularly in terms of depression, has increasingly become a concern. This study aims to analyze the use of Artificial Intelligence (AI) models, specifically Natural Language Processing and Deep Learning, in detecting negative emotions among social media users, with a focus on classifying types of negative emotions. The methods used in this study involve Machine Learning approaches, namely Support Vector Machine (SVM), Extreme Gradient Boosting (XGBoost), Light Gradient Boosting Machine (LIGHT GBM), and Categorical Boosting (CatBoost). Other methods used in this study also involve a Deep Learning approach, namely Bidirectional Encoder Representations from Transformers (BERT), Decoding-enhanced BERT with Disentangled Attention (DistilBERT), Robustly Optimized BERT Approach (RoBERTa), and Decoding-enhanced BERT with Disentangled Attention (DeBERTa), which excel at automatically identifying emotions from text uploaded by users. The dataset used is DepressionEmo, which is designed to detect 8 emotions related to depression, such as Anger, Cognitive dysfunction, Emptiness, Hopelessness, Loneliness, Sadness, Suicide intent, and Worthlessness, through 6037 examples of long text posts by Reddit users. The results of this research are expected to produce an effective model for classifying emotion types based on text, provide deeper insights into negative emotion types, and offer practical solutions in the development of AI models for emotion detection, as well as evaluate AI models in emotion detection. The best Deep Learning variation used was microsoft/deberta-base with hyperparameters consisting of hidden layers valued at 12, batch size valued at 8, learning rate valued at 2e-5, and Adam optimizer type, which obtained a micro F1 score of 0,8254. A comparison of Machine Learning methods also showed that the deep learning model provided better values on the F1 Micro Score evaluation metric.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Media Sosial, Depresi, Machine Learning, Deep Learning, Klasifikasi Emosi Negative, Analisis teks, Multilabel, Social Media, Depression, Negative Emotion Classification, Text Analysis
Subjects:	Q Science > Q Science (General) > Q325.5 Machine learning. Support vector machines. Q Science > QA Mathematics > QA336 Artificial Intelligence Q Science > QA Mathematics > QA76.87 Neural networks (Computer Science) T Technology > T Technology (General) > T57.5 Data Processing
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Informatics Engineering > 55201-(S1) Undergraduate Thesis
Depositing User:	Fauzi Rizki Pratama
Date Deposited:	31 Jul 2025 02:12
Last Modified:	31 Jul 2025 02:12
URI:	http://repository.its.ac.id/id/eprint/123434

Actions (login required)

View Item